M. Herceg and C.C.Tscherning, University of Copenhagen

 

Evaluation of Least-Squares Collocation and

the Reduced Point Mass method using the International Association of Geodesy,

Joint Study Group 0.3 test data.

EGU 2014

Least-Squares collocation (LSC) and Reduced point masses (RPM)

EGU 2014

• Both methods use radial base functions for constructing approximations to T=W-U:

• LSC: Reproducing Kernels (in all obs. Points)

• RPM: Reduced point mass potentials in grid• M

Approximation of anomalous potential, .

EGU 2014

• , harmonic function = linear combination of base-functions on which the observation functional has been applied wrt. Q.

• Requires global data-coverage, but JSG 0.3 data are regional, so egm2008 coefficients up to degree N used as observations.

• Equivalent to EGM is subtracted and later added. EGM96 error-degree variances used to represent the error (arbitrary choice).

Reproducing kernel determination / Covariance fitting:

EGU 2014

• Covariance functions with N=241 and N=37 estimated in the two test areas and used to determine analytic representation as a reproducing kernel.

• Fitting difficult in Pacific area due to extreme values. (In practice, residual topographic effects would have been used to smooth values).

Empirical covariances and Reproducing kernel model:

EGU 2014

RPM grid and depth selection:

EGU 2014

• For RPM we have to fix the position

• Grid spacing: 0.25° x 0.50°

• Depth of the sources (Bjerhammer sphere) is 20km

Calculations/estimations of :

EGU 2014

• Low and high resolution ground data used (11335 values).

• Airborne data used at altitude (as only source)• ”GOCE” data used at satellite positions• ”GRACE” δT values used at satellite position• Ground computed from ”GRACE” and

”GOCE” data – results not shown.• Detailled results available at

http://cct.gfy.ku.dk/jsg03.htm

Results: Differences prediction from of – T “observed”,

EGU 2014

Europe LSC RPM Pacific LSC RPM

”obs” Diff Error Diff ”obs” Diff Error Diff

Mean -0.01 -0.03 0.80 0.07 -0.13 -0.06 0.79 -0.40

St.dev 4.08 0.82 1.44 4.82 0.24 2.94

Results: Differences prediction from of – T “observed”,

EGU 2014

Difference

(Obs - RPM prediction)

Difference

(Obs - Col prediction)

Observations (contribution up do d/o 240 is subtracted)

Differences prediction of ):

EGU 2014

From Airborne , EGM2008 to 240 subtracted:

Europe (LSC) Pacific (LSC)

Obs Dif Err Obs Diff Err

Mean -001 -0.11 2.97 -0.13 -0.08 1.65

St.dev 4.08 2.73 4.82 1.92

Differences prediction of ):

EGU 2014

From GOCE , and EGM08 to 36:Europe (LSC) RPM Pacific (LSC) RPM

Obs Diff Err Diff Obs Diff Err Diff

Mean -0.11 -0.26 6.80 0.06 4.62 -0.18 8.41 -0.74

St.dev 32.55 6.03 5.78 49.07 6.66 6.92

Difference

(Obs - RPM prediction)

Difference

(Obs - Col prediction)

Differences prediction of ):

EGU 2014

From GRACE potential differences, and EGM08 to

36:Europe (LSC) Pacific (LSC)

Obs Dif Err Obs Diff Err

Mean -0.11 -1.76 28.75 4.62 0.19 24.64

St.dev 32.55 19.15 49.07 18.75

Conclusion (1)

EGU 2014

• Good agreement between differences and error-estimates for LSC. Errors large at borders to lower resolution data.

• Results unbiased considering error estimates.• Good agreement for LSC and RPM• Results in Europe of 8 cm, Pacific 2 cm excellent, but

Pacific error-estimate larger.

Conclusion (2)

EGU 2014

• Improvement in results if Topography or observations of EGM08 coefficients to higher degree was used (JSG decision)

• RPM must be further developed

1. in order to use potential differences (GRACE) defined by JSG 0.3.

2. to account for errors in EGM used.

• RPM experiments with grid point selection needed.